Comparison of Clinical Prediction Rules for Ruling Out Cirrhosis in Nonalcoholic Fatty Liver Disease (NAFLD)

Danielle Brandman; Marie Boyle; Stuart McPherson; Mark L. Van Natta; Arun J. Sanyal; Kris Kowdley; Brent Neuschwander-Tetri; Naga Chalasani; Manal F. Abdelmalek; Norah A. Terrault; Art McCullough; Ricki Bettencourt; Cyrielle Caussy; David E. Kleiner; Cynthia Behling; James Tonascia; Quentin M. Anstee; Rohit Loomba

Disclosures

Aliment Pharmacol Ther. 2022;55(11):1441-1451. 

In This Article

Discussion

This study leverages data from two large multicentre, multinational studies, to evaluate the diagnostic performance of seven clinical prediction rules to rule out cirrhosis in patients with biopsy-confirmed NAFLD in well-characterised, predominantly white, <65-year-old patient cohorts that included mostly patients who were white, female and younger than 65. Among the seven different clinical prediction rules evaluated, FIB-4 and the NAFLD fibrosis score had excellent negative predictive value to exclude cirrhosis. The high sensitivity and NPV of both scores indicate that few patients would be misclassified as not having cirrhosis despite its presence on liver biopsy. These results also demonstrate that the previously well-validated cutoffs for ruling out advanced fibrosis perform well for ruling out cirrhosis in this patient population. We propose the use of new cutoffs for FIB-4 (1.67) or the NAFLD Fibrosis Score (0.28) to identify NAFLD patients with similar demographic characteristics who are unlikely to have cirrhosis have similar demographics as in our cohorts. This method facilitates the identification of patients at low risk for cirrhosis who may not need subspecialty care and helps to preserve access for patients with cirrhosis to hepatologists, with the aim of reducing liver-related morbidity and mortality. Because the cohort lacks racial and ethnic diversity within the US, caution should be used in applying these novel cutoffs in non-white patients. Earlier studies have demonstrated that higher cutoff values are likely appropriate for older patients,[34] so further validation of our novel cutoffs may be required prior to application in patients older than 65 to ensure appropriate sensitivity is preserved.

While prior studies have not specifically set out to identify those who may have cirrhosis, these studies have had similar results to ours with regard to the performance of clinical prediction rules in diagnosing advanced fibrosis. Our results are similar to what was observed in a meta-analysis of four studies that included a total of 1038 patients, where the NPV to rule out advanced fibrosis (F3-4) for FIB-4 and the NAFLD fibrosis scores were 0.9–0.91.[35] Moreover, our new FIB-4 and NAFLD fibrosis score cutoffs to rule out cirrhosis maintain a similarly high level of accuracy for identification of cirrhosis.[16] These results reaffirm the use of FIB-4 and NAFLD fibrosis score cutoffs to exclude the presence of cirrhosis in NAFLD patients and identify those who are at high risk but may need further testing to confirm the diagnosis.

Our use of the Youden index relied upon a high sensitivity for each clinical prediction rule (90%), with the goal of yielding robust NPV to rule out cirrhosis. Our methods yielded a single cutoff that may better inform primary care providers about which of their fatty liver patients are indeed at low risk for advanced fibrosis. This cutoff selection method also assumed the risk of classifying a patient as not having cirrhosis was the same as classifying a patient as having cirrhosis, which is not necessarily true clinically. While a false-negative test may eliminate opportunities to reduce liver-related morbidity and mortality through HCC and/or variceal surveillance,[36] a false positive may lead to additional unnecessary testing and cause undue anxiety to the patient. These risks have to be considered in the context of the populations where the tests are used. Since the prevalence of cirrhosis is low in the general population, the high negative predictive value of these new cutoffs is likely to represent true negatives. The low specificity and positive predictive value of these risk scores further underscore the need for additional approaches to confirm the presence of cirrhosis if patients are suspected to have cirrhosis on the basis of noninvasive tests. Use additional testing (eg, transient elastography, ELF, liver biopsy) will be necessary to confirm the presence of cirrhosis in patients with FIB-4 and NAFLD Fibrosis Score values above these cutoffs.

Although the specificity for the new cutoffs proposed for FIB-4 and NAFLD Fibrosis Score for cirrhosis in this current study is relatively low, the high sensitivity and NPV minimise the risk that patients will pay the price of misclassification into lower-risk groups. While we aimed to identify patients with cirrhosis, given the suboptimal specificity and overlap of these new cutoffs for FIB-4 and NAFLD fibrosis scores with those previously established for advanced fibrosis (F3), it is likely that some patients with advanced fibrosis will be labelled as having cirrhosis. Given the high risk of liver-related events in patients with advanced fibrosis, this potential "misclassification" may result in appropriate linkage to subspecialty care. The other "misclassification" applies to the few patients with cirrhosis but are assessed as not having cirrhosis due to scores below the cutoffs. Future studies are needed to determine if repeatedly calculating FIB-4 scores and the NFS at routine follow-up visits will eventually identify those patients.

This study does have some limitations that must be acknowledged and may limit the generalisability of our findings. First, our cohort comprised a broad spectrum of NAFLD severity, which parallels that seen in the general NAFLD population. Even though the cohorts contain a relatively small number of patients with cirrhosis, it is likely that even this proportion is greater than that in the general population as both cohorts were compiled at highly specialised centres. If spectrum bias is present, this could affect the performance of the clinical prediction rules evaluated and their respective derived cutoffs. While the new cutoffs for cirrhosis for FIB-4 had excellent performance in the UK cohort, the lack of validation in more ethnically diverse cohorts is needed, particularly since the negative predictive value of FIB-4 will decrease in populations where the prevalence of cirrhosis is higher.[38,37] The gender of included patients was predominantly female, which may also affect the performance of FIB-4. Additionally, the cohorts used for this study had only a small minority of patients with age >65 years and thus the performance of these cutoffs could not be further assessed in the older population. Finally, our cross-sectional study design does not account for variability in lab measurements over time. Further studies are needed to assess the trend in FIB-4 and its association with fibrosis progression.

It is impractical to subject the 25% of the US population at risk for NAFLD[1] to liver biopsy to accurately characterise disease severity, particularly since a very small proportion of this large group will actually have cirrhosis, yet it is important to identify them for closer monitoring. Moreover, clinical prediction rules that rely on complex formulas such as a previous score developed using the NASH CRN cohort[39] often do not gain widespread acceptance in routine clinical care. Application of simple clinical prediction rules with a single cutpoint may allow primary care providers to identify patients who may be at lower risk for having cirrhosis due to NAFLD on a large scale, thereby maintaining the capacity for higher-risk patients to be referred to gastroenterologists and hepatologists for further risk stratification. We have shown that simple methods utilising readily available clinical data accurately rule out cirrhosis in patients with NAFLD. These methods can be applied easily in the primary care setting to white, younger patients to avoid speciality referral for low-risk patients, though validation in populations who are older, non-white and male is needed.

processing....